Reimbursements · from a YouTube video, Sep 26, 2026 · Added
FBA Problem Finder Across Seven Inventory and Reimbursement Reports with Codex
Cross-check inventory, reimbursement, return, removal and ledger reports to find the few cases worth chasing
“It went through 89,000 records and finally narrowed everything down to just four cases”
Per source Self-reported by the source, not independently verified.
Copy-paste prompt
I'm giving you these FBA reports: inventory ledger, reimbursements, customer returns, removal orders, removal shipments, inventory adjustments and current inventory. Join them by SKU/FNSKU and date and find units that were lost, damaged, returned or removed without a matching reimbursement or reconciliation. Write a script for the math rather than estimating. Output one row per case: SKU, what happened, units, dates, evidence rows, and a label (act now / investigate / monitor / no action). Then build a simple dashboard of the results. Don't file anything.
The source didn't publish its exact prompt. We wrote this starter so you can try the same workflow. Replace anything in [brackets].
See the original YouTube video
What they did
Shikha Mishra gave Codex seven FBA reports and asked it to cross-analyze them, connecting inventory movements, reimbursements, returns, removals, ledger data and current inventory instead of reading one report at a time. A small decision model from a separate API she was testing then labeled each flagged case as act now, investigate, monitor or no action, and Codex turned the whole analysis into a dashboard.
How it went
"Codex did the heavy lifting. It went through 89,000 records and finally narrowed everything down to just four cases that actually needed attention." The decision step used three requests, around 2,700 tokens. Everything ends up on one dashboard instead of seven reports.
We only report what the source says. Results are self-reported unless noted.
Worth knowing
A 2-minute demo. The video doesn't say whose account the reports came from or whether the four cases were filed or paid. The decision API is only partly named in the video, so it isn't tagged here.